Key Takeaways

  • Stranded power plants occur when an industrial customer closes down, leaving a fully capitalized generator with zero local buyers for its electricity.
  • Bitcoin and Ethereum miners solved this by moving modular data centers directly to remote generation sites, trading electricity costs against hardware depreciation.
  • CoreWeave built data centers to rent GPUs to Ethereum miners, securing long-term power grid delivery commitments in the process.
  • When machine learning demand exploded, CoreWeave swapped crypto rigs for Nvidia H100s inside the same physical facilities.
  • AI data centers must operate as continuous baseload consumers because the rapid depreciation of expensive chips penalizes downtime far more than electricity costs.

Stranded Generation Needs Mobile Off-Takers

Power plants require massive up-front capital investment. Once a plant is built, generating electricity only requires operating costs like fuel and maintenance. Problems arise when a power plant is built specifically to supply a co-located industrial facility, such as a paper mill or smelter. If that industrial customer goes bankrupt or relocates, the power plant is suddenly stranded. The capital expenditure is already spent, yet the plant cannot sell its electricity without expensive, slow-to-build transmission connections to distant cities.

As Patrick McKenzie explains: “The capex, heavy in most generation businesses, has been paid, and if you put in opex you will get electricity out, but useless without a local-ish consumer of a large block of electricity.”

Crypto miners emerged as the ideal buyer for these stranded assets. Mining hardware requires only power and an internet uplink to monetize electricity immediately. McKenzie notes that crypto operations pit hardware depreciation against power tariffs: “Bitcoin miners are in the business of pitting electronics depreciation and electricity consumption against each other to win a tournament in generating random numbers with mystical properties.” Because crypto miners could deploy anywhere in large capital increments, they moved straight to the cheapest stranded generation sites they could find.

The Real Asset Was the Interconnect

CoreWeave started inside this crypto hosting market. The company set out to buy GPUs and rent them to Ethereum miners. As McKenzie recalls: “CoreWeave originally started renting out I believe GPUs to Ethereum miners with the idea that Ethereum was also going to take off and become a large business in the vein of Bitcoin as some sort of global worldwide computer.”

Crypto mining fluctuates wildly, but building those facilities forced CoreWeave to solve the hardest bottleneck in physical infrastructure: securing transmission interconnects and multi-year supply contracts from power utilities. When enterprise demand shifted toward machine learning, those utility contracts became far more valuable than the crypto tokens they originally mined.

McKenzie describes the pivot plainly: "If you have a data center which is built to host GPUs for the purpose of Ethereum mining, what you have is a powered box and you often have a multi-year commitment from the power grid to continue providing power for the box, and so you can stuff that box full of GPUs doing anything else, like for example H100s doing inference or training runs for AI companies, and that turns out to be a much more valuable global computer than Ethereum was."

This created a distinct operating model. Bitcoin miners like Riot can turn off their ASICs during peak demand to sell power back to the grid. AI workloads cannot do that. An Nvidia H100 depreciates rapidly over three years. Leaving a cluster idle to save on afternoon electricity rates burns more capital in chip value than it saves on power bills. AI clusters operate as continuous baseload consumers, meaning that guaranteed, uninterrupted grid interconnects are the scarce resource controlling the sector.

What to Do With This

Audit your company's dependencies for long-lead physical bottlenecks. If you are building compute-heavy products, stop negotiating solely with software vendors and start identifying facilities with active utility interconnection agreements. Sign multi-year access agreements for power capacity before ordering high-end hardware, because acquiring compute takes weeks while getting a new grid connection approved takes years.